Again: near singular matrix

[gerrard]

Hi,

I know this question has been posed a lot of times before, but the thing is that I cannot figure out which variables are colinear in my regressions.

I tried to estimate these two equations:
roa1_win c ind1 size1_win age2_win gr1_win risk1_win herf shceo tan_win list2 med_lev1
Sample: 2000 2012 if lev1_win>=0.25486 and lev1_win<0.65246 and fam=1

lev1_win c ind1 size1_win age2_win gr1_win risk1_win herf shceo bos tan_win list2 med_lev1
Sample: 2000 2012 if lev1_win>=0.25486 and lev1_win<0.65246 and fam=1

For both, I get "Near singular matrix". None of the variables are dummy variables, so no dummy variable trap. In the sample restriction though, fam is a dummy variable. But I need this specification because I only want to consider family firms (fam=1)
I tried excluding herf and shceo, since they are highly correlated, but that made no difference. Same with excluding bos, as correlation was quite high with herf.

In any case, excluding any one of those will be difficult, as this will make my simultaneous equation system (I', using 2SLS, those above are the reduced form equations) unidentified.

Please help me figuring out how I can solve this.
Thanks!

[startz]

The first step, which you may have already done, is to set the sample exactly as it's set in your estimation and then to look at the correlation matrix of all your RHS variables.

[gerrard]

Ah I have found the reason!
LIST2 is NA if fam=1 (no family firm in my sample is listed).

Excluding it solves the problem. But does it induce an identification problem in my two equation system?

My structural equations will be (w/o LIST2)

1) roa1_win c lev1_win (1/2)*lev1_win^2 ind1 size1_win age2_win gr1_win risk1_win herf shceo bos
2) lev1_win c roa1_win (1/2)*roa1_win^2 ind1 size1_win age2_win gr1_win risk1_win tan_win med_lev1

Is the system/are the equations still identified?